Score for human Activity Recognition?

Hello,

I recently try to start a project according to the excellent link below, I call it example project.

In my new project, I plan to collect human activity movement data by some “expert” people to build a “expert” model and save it.
So, for example, when I do some activities, the expert model can not only predict what activity I am doing, and tells me the “Similarity”.
It means I can realize how much difference between me and the experts in the same activity movement.

Can anyone tells me if my idea workable? or give me any hint?
Are there any methods I can use in TensorFlow?

Any suggestions will be appreciated.
Thanks sincerely…

Hi Anthony,

This is a very interesting idea. Some parts that might be hard:

  • Have the comparison (pro vs user) be on the same perspective and position (rotation wise)

You might need a model (or algorithm) to pair the movement execution and then compare the differences.

There’s also some discussion on the topic here: https://discuss.tensorflow.org/t/what-model-s-for-a-sequence-of-human-poses

Thanks so much for your reply.

1 Like